Amodei says open source in AI doesn't work like in software—since we can’t see inside models, it’s more like “open weights.” Key benefits of open source—collaborative development, transparency—don’t translate well to large model inference. He prioritizes whether a model performs well, not whether it’s open or closed: “I don’t care if it’s open source, I care if it’s good.” Cloud access, fine-tuning, and interpretability tools are more relevant than raw openness in evaluating AI competition.